COMPARING MODEL SIMULATIONS OF THREE BENCHMARK TSUNAMI ...

Grilli

Description

COMPARING MODEL SIMULATIONS OF THREE BENCHMARK
TSUNAMI GENERATION CASES
1 2 3By Philip Watts , Fumihiko Imamura and Stéphan Grilli
1Applied Fluids Engineering, PMB #237, 5710 E. 7th Street, Long Beach, CA 90803.
2Prof., Disaster Control Research Center, School of Engrg., Tohoku University, Aoba 06,
Sendai 980-8579, Japan.
3Prof., Dept. Ocean Engineering, University of Rhode Island, Narragansett, RI 02882.
KEYWORDS: Tsunami, tidal wave, wave generation, benchmark case, simulation,
underwater landslide, submarine landslide, sub-aqueous landslide
ABSTRACT: Three benchmark cases are proposed to study tsunamis generated by
underwater landslides. Two distinct numerical models are applied to each benchmark case.
Each model involves distinct center of mass motions and rates of landslide deformation.
Computed tsunami amplitudes agree reasonably well for both models, although there are
differences that remain to be explained. One of the benchmark cases is compared to
laboratory experiments. The agreement is quite good with the models. Other researchers
are encouraged to employ these benchmark cases, in future experimental or numerical
work.
INTRODUCTION
Tsunamis generated by underwater landslides are receiving more attention following
analyses demonstrating that the surprisingly large local tsunami documented during the
1998 Papua New Guinea catastrophe was generated by submarine mass failure (Kawata et
al., 1999; Tappin et al., 1999, 2000; Synolakis et al., 2000). In ...

Sujets

• Modelica
• Séisme de 2004 dans l'océan Indien
• Amplitude
• Gwerz
• EQUIDYN
• Submarine Rescue Diving and Recompression System

Informations

• Publié par: Ondroi
• Nombre de lectures: 44
• Langue: English

Extrait

COMPARING MODEL SIMULATIONS OF THREE BENCHMARK TSUNAMI GENERATION CASES By Philip Watts 1 , Fumihiko Imamura 2   and Stéphan Grilli 3 1 Applied Fluids Engineering, PMB #237, 5710 E. 7th Street, Long Beach, CA 90803. 2 Prof., Disaster Control Research Center, School of Engrg., Tohoku University, Aoba 06, Sendai 980-8579, Japan. 3 Prof., Dept. Ocean Engineering, University of Rhode Island, Narragansett, RI 02882. KEYWORDS: Tsunami, tidal wave, wave generation, benchmark case, simulation, underwater landslide, submarine landslide, sub-aqueous landslide ABSTRACT:  Three benchmark cases are proposed to study tsunamis generated by underwater landslides. Two distinct numerical models are applied to each benchmark case. Each model involves distinct center of mass motions and rates of landslide deformation. Computed tsunami amplitudes agree reasonably well for both models, although there are differences that remain to be explained. One of the benchmark cases is compared to laboratory experiments. The agreement is quite good with the models. Other researchers are encouraged to employ these benchmark cases, in future experimental or numerical work. INTRODUCTION Tsunamis generated by underwater landslides are receiving more attention following analyses demonstrating that the surprisingly large local tsunami documented during the 1998 Papua New Guinea catastrophe was generated by submarine mass failure (Kawata et al., 1999; Tappin et al., 1999, 2000; Synolakis et al., 2000). In response to these and other studies, recent work by marine geologists now considers the tsunamigenic potential of submarine mass failure scars (Goldfinger et al., 2000; Driscoll et al., 2000). Despite these advances in the observational science, there remains no validation of the numerical models currently in use. Consequently, the ability of scientists to simulate tsunami generation by underwater landslides remains in doubt. Note, underwater landslides are also called submarine mass failures and display a wide range of morphological features (Prior and Coleman, 1979; Edgers and Karlsrud, 1982; Hampton et al., 1996). 1

Researchers have tackled tsunami generation by underwater landslides with a wide variety of numerical methods incorporating many different assumptions. Iwasaki (1987, 1997) and Verriere and Lenoir (1992) utilized linear potential theory to simulate wave generation by moving the domain boundary. Depth-averaged Nonlinear Shallow Water (NSW) wave equations were solved by Fine et al. (1998), Harbitz (1992), Imamura and Gica (1996), and Jiang and LeBlond (1992, 1993, 1994), in combination with disparate landslide models. Fully nonlinear fluid dynamic field equations were solved by Assier Rzadkiewicz et al. (1997), Grilli and Watts (1999) (in an irrotational and inviscid approximation), and Heinrich (1992), in concert with assorted landslide models. Watts et al. (2000) appears to be the only work to compare tsunami generation for different center of mass motions and different rates of landslide deformation. For the most part, scientists have studied vastly different landslide geometries, motions, and constitutive behaviors. There is currently no consensus on the ability of these different models to reproduce tsunami generation by underwater landslides. To further advance research in tsunami generation by underwater landslides, benchmark cases are needed to validate numerical models and to help explain the origins of any discrepancies that may exist, both between numerical models, and with comparisons to experimental results. Benchmark problems are already available for tsunami propagation and inundation (Liu et al., 1991; Yeh et al., 1996). Our goal in this work is to establish three benchmark cases for future reference by researchers interested in tsunami generation by underwater landslides, and to compare simulations, for one of these cases, to recently performed laboratory experiments. Each case is two-dimensional in order to reduce computational or experimental effort.   z x g d T b s)t(
x Figure 1: Definition sketch of the simulation domain in II and GW Models, and of initial landslide parameters 2

We compare results from two distinct numerical models. We hope that this work will promote future numerical and experimental comparisons. The comparisons made here are by no means the end of this effort. BENCHMARK CASES To facilitate their experimental realization, the benchmark cases chosen for this work are based in part on the sliding block experiments of previous researchers (Heinrich, 1992; Iwasaki, 1982; Watts, 1997; Wiegel, 1955). A straight incline forms a planar beach with the coordinate origin at the undisturbed beach and the positive x-axis oriented horizontally away from the shoreline (Fig. 1). A semi-ellipse approximates the initial landslide geometry. Landslide deformation is permitted following incipient motion of the semi- ellipse. The nominal underwater landslide length measured along the incline is b =1000 m for all three cases. All underwater landslides are assumed to have a bulk density  b  =1900 kg/m 3  and fail in sea water of density  o  =1030 kg/m 3 . The geometrical parameters for each benchmark case are given in Table 1. The initial submergence at the middle of the landslide, x = x g , was obtained from a scaled reference equation d = b sin
, while the initial landslide thickness was calculated from another scaled reference equation, T = 0.2 b sin
   (Watts et al., 2000). A wave gage was situated above the middle of the initial landslide position at x g = (d + T/ cos
)/ tan
, and recorded tsunami elevation  (t). Dimensional quantities are presented throughout since different numerical techniques employ different non-dimensional schemes. Watts (1998) provides the correct Froude scaling to perform these benchmark experiments at laboratory scale. Table 1: Underwater landslide and numerical wave gage parameters for benchmark cases c 1, c 2, and   c 3
bTd x g Case(m)(m)(m)(m) (1)(2)(3)(4)(5)(6) c 1 30˚10001005001066 c 2 15˚100051.82591166 c 3 5˚100017.487.21196 3

LABORATORY EXPERIMENTS Laboratory experiments were conducted in the University of Rhode Island wavetank (length 30 m, width 3.6 m, depth 1.8 m). This tank is equipped with a modular beach made of 8 independently adjustable panels (3.6 m by 2.4 m) whose difference in slope can be up to 15 o . Benchmark case 2 was tested in the wave tank at 1:1000 scale, in the set-up shown in Fig. 2. Two beach panels were set to an angle
 =15 o and covered by a smooth aluminium plate. A quasi two-dimensional experiment was realized by building vertical (plywood) side walls at a small distance (about 15 cm) from each other. A semi-elliptical wood and plastic landslide model was built and installed in between the walls. The model was equipped with low-friction wheels and a lead ballast was added to achieve the correct bulk density (Fig. 3). An accelerometer was attached to the model center of gravity to measure landslide kinematics. Four capacitance wave gages were mounted on an overhead carriage, to measure free surface elevation (Fig. 2), the first gage being located at x = x g and the others mounted 30 cm apart with increasing x-positions. Figure 2: Quasi two-dimensional landslide experiments for benchmark case 2 4

Figure 3: Close-up of scale model for two-dimensional landslide experiments Experiments were repeated at least five times and the repeatability of results was very good. Results are presented in a following section. NUMERICAL MODEL DESCRIPTIONS Imamura and Imteaz (1995) developed a mathematical model for a two-layer flow along a non-horizontal bottom. Conservation of mass and momentum equations were depth- integrated in each layer, and nonlinear kinematic and dynamic conditions were specified at the free surface and at the interface between fluids. Both fluids had uniform densities and were immiscible. Vertical velocity distributions were assumed within each fluid layer. The landslide fluid was ascribed a uniform viscosity, which sensitivity analyses show has very little effect on wave records over a range of viscosities 1-100 times that of water. A staggered leap-frog finite difference scheme, with a second-order truncation error was used to solve the governing equations. Landslides were thus modeled as immiscible fluid flows comprising a second layer, as in the work of Jiang and LeBlond (1992, 1993, 1994). An instantaneous local force balance governed landslide motion. Hence, this motion resulted from the solution of the problem itself and was not externally specified as a boundary condition. We will refer to this numerical model as the II Model below. Grilli et al. (1989, 1996) developed and validated a two-dimensional Boundary Element Model (BEM) of inviscid, irrotational free surface flows (i.e., potential flow theory). Cubic boundary elements were used for the discretization of boundary geometry, combined with fully nonlinear boundary conditions and second-order accurate time updating of free surface position. The model was experimentally validated for long wave propagation and runup or breaking over slopes by Grilli et al. (1994, 1998). Model predictions are 5

surprisingly accurate; for instance, the maximum discrepancy for solitary waves shoaling over slopes is 2% at the breaking point, between computed and measured wave shapes. Grilli and Watts (1999) applied this BEM model to water wave generation by underwater landslides and performed a sensitivity analysis for one underwater landslide scenario. The landslide center of mass motion along the incline was prescribed by the analytical solutions of Watts (1998, 2000) (see next section). In these computations, the landslide retained its semi-elliptic shape while translating along the incline. We will refer to this numerical model as the GW Model below. )m( s 0002 0051 0001 005 c 1 c c 2 1 c 2 c 3 c 3 0 t (s) 0102030405060 Figure 4: Underwater landslide center of mass motion as a function of time in the II (solid) and GW (dashed) Models, for benchmark cases c 1 , c 2 , and c 3  in Table 2 Both the II and GW Models are used in the following to simulate tsunamis generated by underwater landslides of identical initial characteristics corresponding to the three benchmark cases in Table 1. For discretization techniques and numerical parameters used in both models, please refer to Imamura and Imteaz (1995) and Grilli and Watts (1999). SIMULATION RESULTS Descriptions of tsunami generation by underwater landslides should begin by documenting landslide center of mass motion and rates of deformation. Since both motion and deformation were prescribed in the GW Model, we proceed to describe the results obtained 6

from the II Model and compare these results with the GW Model. We also relate the measured initial acceleration obtained for case 2. Assuming the center of mass motion s(t) is parallel to the incline (Fig. 1), Fig. 4 shows the center of mass motions obtained in the II Model for the three benchmark cases. It is readily verified that the simple equation 2 t a o s(t) =    2 )1( provides an accurate fit of these motions. Eq. (1) is the first term in a Taylor series expansion of landslide motion beginning at rest (Watts, 2000). In fact, two-parameter curve fits of the equation of motion given in Watts (1998) (and reproduced as Eq. (3) below) failed to produce unique parameter values, due to the accuracy of the one-parameter fit given by Eq. (1). Two curve fitting parameters introduced a redundancy in the solution algorithm that yielded infinite fitted solutions. Values of initial landslide accelerations a o for the II Model obtained by curve fitting Eq. (1) can be found in Table 2. Note that R 2 coefficients were 0.99 or better for all of the fits. The experimental initial acceleration was a o  = 0.73 m/s 2  for case 2. This compares favorably with the value from the GW Model in Table 2 and suggests an added mass coefficient C m   1.2 given negligible rolling friction. Table 2: Initial accelerations, terminal velocity and rates of deformation in II and GW Models a oII a oGW u tGW  II  GW Case(m/s 2 )(m/s 2 )(m/s)(s -1 )(s -1 ) (1)(2)(3)(4)(5)(6) c 1 3.111.4780.90.0620.000 c 2 1.290.7657.80.0350.000 c 3 0.400.2633.20.0170.000 Landslide deformation in the II Model was manifested foremost as an extension in time, b(t), of the initial landslide length b o . Fig. 5 demonstrates that the non-dimensional ratio b/b o  varies almost linearly with time, following an initial transient, similar to the experimental observations made by Watts (1997) for a submerged granular mass. A semi- empirical expression that describes landslide extension is 7

b(t)  = b o  {1 +   t [1 - exp(-K t )]}(2) where    is the eventual linear rate of extension and the exponential term describes an initial transient, with K = a o /g    (Watts et al., 2000). The parameter K is chosen to fix the uppermost landslide corner in place as the center of mass begins to accelerate. Table 2 gives values of   for the II Model found from curve fits of Eq. (2). b/b o 3 c c 1 2 5.2 2 c 3 5.1 1 t (s) 0102030405060 Figure 5: Underwater landslide temporal extension in II Model Watts (1998) developed a wavemaker formalism for non-deforming underwater landslides, based on an analytical solution of center of mass motion t s(t) = s o  ln [ cosh (   t o     )] (3) htiw 2 uu tt s o   = a o     , t o   = a o   (4a,b) where a o    and u t     denote landslidei nitial acceleration and terminal velocity, respectively (see Eq. (5) and discussion in the following section). Eqs. (3) and (4) were used in the GW Model to specify the landslide kinematics. Eq. (4) can also be expressed as a function of 8

the landslide physical parameters: initial length, incline angle, and density (Watts, 1998). For the three benchmark cases, using the data in Table 1, we find the values of a o    and u t    g  (m) 2 0 2- 4- 6- 8- 01- 21- -14 t (s) 01020304050 Figure 6: Numerical wave gage record atx  g   = 1066 m for benchmark case 1; II Model (solid); GW Model (dashed)  g  (m) 2 0 2- 4- 6- 8- 0102030405060 9 s( t)

Figure 7: Numerical wave gage record at x g   = 1166 m for benchmark case 2; II Model (solid); GW Model (dashed); scaled-up experiments (dots) listed in Table 2 and corresponding motion s(t) shown in Fig. 4. Note, as discussed above, no extension    was specified in the GW Model. Figures 6-8 show the tsunami simulation results of both numerical models for cases 1-3, respectively. The GW and II Model results agree qualitatively for all three cases, although the GW Model produces slightly smaller wave amplitudes. The II Model produces more acute free surface curvature near t = 0 as well as longer tsunami periods. Maximum tsunami amplitudes at the numerical wave gages are given in Table 3. This is the same characteristic tsunami amplitude employed in the scaling analyses of Watts (1998, 2000). Note, the II Model has water wave disturbances in the first 5-20 s of each simulation brought on by a Kelvin-Helmholtz type instability along the landslide-water interface. Table 3: Simulated and calculated characteristic wave amplitudes DISCUSSION  II  GW  PP Case(m)(m)(m) (1)(2)(3)(4) c 1 11.9810.8615.71 c 2 6.226.378.14 c 3 2.072.392.73 Tsunami generation in the shallow water wave limit occurs through vertical acceleration of some region on the ocean floor (Tuck and Hwang, 1972; Watts et al., 2000). Since the center of mass motion modeled in the II Model, as shown in Fig. 4, corresponds to the landslide acceleration described by Eq. (1), tsunami generation by the II Model in Figs. 6-8 can be directly associated with vertical landslide acceleration. Tsunami generation in a potential flow model such as the GW Model, however, occurs through gradients of the velocity potential at the free surface, which can arise from both horizontal and vertical landslide motions. Also, tsunami generation in the GW Model is theoretically not limited to landslide acceleration and may include the instantaneous water velocity distribution. 01

 g  (m) 5.0 0 5.0- 1- 5.1- 2- 5.2- 002040680100 t (s) Figure 8: Numerical wave gage record atx  g   = 1196 m for benchmark case 3; II Model (solid); GW Model (dashed) The initial center of mass motion during landslide tsunami generation can be accurately described by Eq. (1), assuming the correct initial accelerationi s known. Along an infinite incline, an equation such as (3) provides a better description of the motion. Watts (1998) provides an analytical method for choosing between Eqs. (1) and (3) based on the length of the incline. Tsunami amplitudei s scaled by the landslide initial acceleration( Watts, 1998, 2000). The initial accelerations listed in Table 2 differ considerably between the two models, despite identical initial landslide shapes and bulk densities. The theoretical initial acceleration specified in the GW Model is, neglecting Coulomb friction, a o     =   (  1)sin
g C  m )5( in which    represents the landslide specific density and C m   an added mass coefficient. Eq. (5) applies specificallyt o underwater landslides that experience negligible basalf riction due to phenomena such as water injection or liquefaction (Watts et al., 2000). The value C m  = 1 used in the GW Model produces conservative landslide motions. Our experimental 11